Classification of Hydrometeors During a Stratiform Precipitation Event in the Rainy Season of Liupanshan

Abstract

This study conducted a classification analysis of hydrometeor types during a typical stratiform mixed cloud precipitation event in the rainy season using data from the Liupan Mountains micro rain radar power spectra. The primary research findings are as follows: (1) Utilizing the RaProM method synthesizes the information of particle falling velocity, equivalent radar reflection coefficient, particle scale characteristics at different stages, and the location of the bright zone in the zero-degree layer to classify hydrometeors during this precipitation process, and the results show that drizzle and raindrop distribution time periods do not match with the raindrop spectra and rain intensities observed by the DSG5 ground-based precipitation gauge. (2) Sensitivity experiments conducted on the RaProM method revealed that after modifying the discrimination thresholds for drizzle and raindrops, the distributions of drizzle and raindrops were more aligned with ground-based raindrop spectrum observations. Furthermore, these adjustments also showed better consistency with the radar reflectivity factor, Doppler velocity, and velocity spectrum width thresholds used by existing millimeter-wave cloud radars to discriminate between drizzle and raindrops. (3) Various kinds of hydrometeors show different vertical distribution characteristics in three precipitation stages: weak, strong, and weak. In the two weak precipitation stages, hydrometeors mainly existed in the form of snowflakes at altitudes above the zero-degree layer and in the form of drizzle at altitudes below the zero-degree layer. The vertical distribution disparity of hydrometeors between the mountain peak and base sites demonstrates that terrain significantly influences hydrometeors during the precipitation process.

1. Introduction

Airborne hydrometeors, including ice crystals, snowflakes, liquid cloud droplets, graupel, drizzle, and others, serve as primary forms of precipitation within or beneath clouds, constituting the core content of cloud precipitation physics research [1,2,3]. The accurate identification of these hydrometeor types is crucial for precipitation forecasting, studies on weather modification, and cloud physics parameterization [4,5,6]. However, due to the complexity of various physical processes within the atmosphere, the detection of different hydrometeors has long been a significant challenge in cloud physics research [7,8,9].

Millimeter-wave cloud radars play a significant role in detecting clouds and weak precipitation, enabling the effective identification of hydrometeor types through the analysis of detection signals [10,11,12,13]. For example, some researchers have established an algorithm for recognizing the phase state of hydrous formations within clouds by combining information from LiDAR, 8 mm radar, and microwave radiometer [10,14]. It has also been proposed to use the higher-order moments of the radar Doppler spectrum to identify cloud particle types [11,12,15,16]. Additionally, research has utilized spectral skewness in the Doppler spectra of millimeter-wave cloud radars to detect drizzle within clouds [13,17].

Due to the increased sensitivity of radars to hydrometeors with shorter wavelengths, the choice of operating frequency for radars varies for different types of hydrometeors [18,19]. Millimeter-wave cloud radar detects precipitation particles at sub-millimeter scales and above with severe signal attenuation and limited ability to identify and invert precipitation particles [15,20,21]. In particular, the performance of cloud radar is significantly limited in the detection of weak echoes, such as weak precipitation phenomena like drizzle [18]. In contrast, the advent of micro rain radar offers possible solutions for weak precipitation detection. [22,23] Not only that but in terms of quantitative precipitation estimation for weak precipitation, the quantitative estimation of micro rain radar is also considered to be more accurate. Peters et al. [24] showed that micro rain radar is capable of accurately measuring rainfall intensity as small as 0.25 mm/h. In addition, micro rain radar has shown significant advantages in precipitation phase identification. For example, Garcia-Benadi et al. [25] effectively identified rain, snow, mixed-phase particles, drizzle, and hail during a precipitation event in the Eastern Pyrenees using Doppler spectra data from a micro rain radar. Wang Hong et al. [26] utilized this method to study the distribution characteristics of hydrometeors in stratiform cloud precipitation and rain-to-snow transition events in Shandong Province.

This study conducted co-observations using millimeter-wave cloud radars, micro rain radars, and the DSG5 precipitation gauge during a stratiform cloud precipitation event in the Liupanshan Mountains. The hydrometeors during the precipitation process were classified, and the impact of terrain on the phase transitions of hydrometeors was investigated. The research findings provide valuable insights for precipitation forecasting, studies on weather modification, and cloud physics parameterization.

2. Materials and Methods

2.1. Information Note

The study area, the Liupanshan Mountains, is located in the southeastern part of the northwest region, in the southern part of the Ningxia Hui Autonomous Region. It lies on the northwest edge of the southwestern monsoon region and experiences a semi-arid to arid climate, with the rainy season occurring from July to September [27]. The research utilized data from two ground observation sites at the mountain peak and the mountain base in the Liupanshan Mountains. The instruments used included a millimeter-wave cloud radar (CR, Xi’an Huateng Microwave Co., LTD, Xi’an, China), a micro rain radar (MRR, METEK GmbH, Elmshorn, Germany), and a DSG5 ground precipitation sensor(Huayun Shengda Company, Beijing, China). The mountain peak site, Liupanshan site (LP), is located at an altitude of 2842 m (106.12° E, 35.40° N), while the mountain base site, Longde site (LD), is situated at an altitude of 2078 m (35.37° N, 106.07° E). The linear distance between the two sites is 10 km (Figure 1).

The CR operates at a wavelength of 8.57 mm (frequency of 35 GHz) and conducts observations using a vertically fixed scanning method, with a time resolution of 5 s and a spatial resolution of 30 m. The MRR, produced by the German company Metek (Meteorologische Messtechnik GmbH, Elmshorn, German), is a vertically pointing continuous wave frequency-modulated Doppler radar with a wavelength of 1.238 cm (Ku-band), a power of 24 GHz, a vertical resolution of 150 m, a time resolution of 60 s, and a maximum measurement altitude of 4.65 km above ground level. The data in this study have a time resolution of 1 min and a height resolution of 150 m. For detailed descriptions of these two instruments, refer to He et al. [28].

The DSG5 Precipitation Phenomenon Gauge has a sampling area of 54 cm² (18 cm × 3 cm) and a time resolution of minutes. The instrument outputs the laser voltage generated by the precipitation particles after shading, thus corresponding to the particle size (diameter range of 0.062 to 24.5 mm) and falling speed (speed range of 0.052 to 20.8 m·s⁻¹) of the precipitation particles and obtains the number of particles in 32 × 32 non-equally spaced classes. The particles within the first two diameter bins were eliminated from the practical study, considering that they had lower signal-to-noise ratios and larger errors. At the same time, raindrops in the real atmosphere will be broken after growing to a certain size, and unrealistically large raindrops in the observation data are mostly caused by the overlap of raindrops when they pass through the sampling area, so raindrops with a diameter larger than 6 mm were also excluded. Raindrops in nature generally do not fall at speeds greater than 9.8 m/s, so data with speeds greater than 9.8 m/s were removed. The quality control process for DSG5 observational data is detailed in the reference by Huo et al. [29].

2.2. MRR Hydrometeor Classification Method

Garcia-Benadi et al. [25] established the RaProM algorithm for hydrometeor classification using micro rain radar power spectra, which identifies snow, drizzle, rain, hail, and mixed phases, and applied it to a precipitation event in the Eastern Pyrenees, with good identification results. In this paper, the method is applied to identify the phases of a single stratiform cloud precipitation over Liupan Mountains.

The RaProM algorithm converts the backscattered signal f(n,i) reflected by precipitation particles into a reflectivity spectral density η(n,i), which is then converted into a velocity v spectral density according to Equation (1).

$$\mathsf{\eta }(\mathsf{\nu },\mathrm{i})=\mathsf{\eta }(\mathrm{n},\mathrm{i}){·({\displaystyle \frac{{\mathrm{f}}_{\mathrm{s}\mathrm{a}\mathrm{m}\mathrm{p}\mathrm{l}\mathrm{i}\mathrm{n}\mathrm{g}}}{2·{\mathrm{n}}_{\mathrm{m}\mathrm{a}\mathrm{x}}·{\mathrm{i}}_{\mathrm{m}\mathrm{a}\mathrm{x}}}}·{\displaystyle \frac{\mathsf{\lambda }}{2}})}^{-1}$$

(1)

Here, n represents the numerical values of the Doppler spectrum lines (n = 0, ···, 63), and i denotes the numerical values corresponding to the range bins (i = 1, ···, 32). fsampling denotes the sampling frequency, set at 125 kHz in this context, with nmax being 64 and imax being 32. λ signifies the wavelength, specified as 1.24 cm in this case.

The algorithms include noise level determination, peak signal detection, and Doppler velocity blurring. For noise level determination in the Doppler spectrum, reference is made to the method proposed by Hildebrand and Sekhon [30], which separates noise from the useful signal. On this basis, peak signal detection is performed, which averages the spectrum to ensure that at least 50 per cent of the spectrum contains a minimum valid signal. On the Doppler velocity blur processing, the falling velocity range of precipitation particles is extended from 0–12 m/s to the deblurring range of −12–24 m/s by utilizing the spectral information of a given altitude layer and neighboring altitude layers. At the same time, the vertical continuity of the velocity profiles is used to guide the deblurring process to provide a reliable velocity spectrum.

Finally, the equivalent radar reflectivity factor $Z_e$ (mm⁶·m⁻³), particle falling velocity ($\overline{w}$), velocity spectral width ($V_w$), and skewness ($S_k$) are calculated using the processed power spectral information. $\overline{w}$, $V_w$, and $S_k$ are the first-, second-, and third-order normalized moments of the sample, respectively, as follows:

$$Z_e={10}^{18}·{\displaystyle \frac{{\lambda }^4}{{\pi }^5}}·{\displaystyle \frac{1}{{\left|K\right|}^2}}·\Delta \nu ·\sum \eta (\nu )$$

(2)

$$\overline{w}={\displaystyle \frac{\sum \eta (\nu ,i)·\nu (i)}{\sum \eta (\nu ,i)}}$$

(3)

$$\mathsf{\sigma }=\sqrt{{\displaystyle \frac{\sum \eta (\nu ,i)·{(\nu (i)-w)}^2}{\sum \eta (\nu ,i)}}}$$

(4)

$$S_k={\displaystyle \frac{\sum \eta (\nu ,i)·{(\nu (i)-w)}^3}{\sum \eta (\nu ,i)·{\sigma }^3}}$$

(5)

The type of hydrometeor was identified by combining the covariates of particle falling velocity, Doppler spectral width σ, skewness ($S_k$), and zero-degree layer bright bands (BB) calculated on each altitude layer, and the flowchart of the phase identification is shown in Figure A1. Mixed phases refer to wet snow, rain–snow mix, and shrapnel with one or more of the other phases. $V_{rain}$ (m·s⁻¹) and $V_{snow}$ (m·s⁻¹) are calculated from the empirical Equations (6) and (7) proposed by Atlas et al. [31] to obtain the average velocities, $V_{rain}$ and $V_{snow}$, for rainfall and snowfall for specific values of $Z_e$, as follows:

$$V_{rain}=2.65·{Z_e}^{0.114}$$

(6)

$$V_{snow}=0.817·{Z_e}^{0.063}$$

(7)

In addition, the presence or absence of the zeroth layer bright band was determined according to the algorithm of Cha et al. [32], and the location of the zeroth layer (BB) was identified according to the method of Wang et al. [33].

2.3. Validation of Hydrometeor Classification Results

A comparison of the proportion of drizzle and rain classified by the MRR at different time intervals with ground-level DSG5 rainfall intensity was conducted. If the proportion of drizzle is higher and the proportion of rain is lower, corresponding to lower surface precipitation intensity, the classification of drizzle is reasonable. Conversely, if the proportion of drizzle is lower and the proportion of rain is higher, corresponding to higher surface precipitation intensity, the classification of drizzle is appropriate.

Furthermore, the hydrometeors classified in this study are assimilated with the CR detection signals at the same time and height levels and compared with various hydrometeor signal thresholds based on CR that are widely referenced [10], demonstrating the validity of the hydrometeor classification results. It is important to note that data where MRR has no detection signal but CR does have a detection signal are treated as cloud particle samples [27].

3. Results

3.1. Weather Background

On 12 August 2021, a precipitation event occurred in the Liupanshan region. A weak eastward westerly trough (see Figure 2a) appeared in eastern Ningxia at 500 hPa before the precipitation, and the southwesterly warm and humid airflow in front of the low-pressure trough affected the Liupan Mountains area. At the 700 hPa level, significant wind shear was evident in the southern winds of Ningxia (Figure 2b), corresponding to the center of positive relative vorticity and the region of negative water vapor flux divergence (Figure 2c,d). This indicates that lower-level air convergence led to upward motion, resulting in the precipitation event.

As seen by the millimeter-wave cloud radar and micro rain radar detection data, the precipitation process appeared as bright bands (Figure 3), which are characteristic of stratiform cloud precipitation [32]. According to the method by Wang et al. [33], the bright band was located at an altitude of 4.65–4.85 km. At the bottom of the western slopes of Liupanshan Mountain (LD), precipitation occurred earlier than at the top of Liupanshan Mountain (LP), with stronger precipitation occurring during the hours 18:00–20:00.

3.2. Classification of Hydrometeor Phase

From Figure 3, it is observed that the original signals from the micro rain radar contain a significant amount of noise. Following the method proposed by Garcia-Benadi et al. [25] to eliminate noise and interference signals using the power spectrum of the micro rain radar, corrections were made to the individual dataset. The radar reflectivity factor and fall speed were then recalculated using the power spectrum data (Figure 4). A comparison between Figure 3 and Figure 4 reveals that after correction, the radar reflectivity factor and fall speed in Figure 4 have had a substantial amount of background noise successfully eliminated.

Using the corrected MRR data and following the hydrometeor classification method proposed by Garcia-Benadi et al. [25], hydrometeors during this precipitation event were classified. The classification results are shown in Figure 5a. Above the height of the zero-degree layer of the hydrometeor main state for snow, snowflakes falling to the zero-degree layer began to melt to form a mixture, and below the zero-degree layer of 200–300 m, they completely melted into liquid precipitation; the main type of raindrops, drizzle, accounted for a small percentage of the drizzle.

Shupe et al. [10] have studied the radar detection signal thresholds corresponding to different types of hydrometeors using millimeter-wave cloud radar and aircraft observations, and Zhu et al. [34] summarized the results of the related studies (see

Table 1. Threshold ranges for radar albedo, Doppler velocity, and velocity spectral width used by millimeter-wave cloud radar to discriminate between types of hydrometeors.

Parameter	Eigenvalue	Snow	Ice	Mixed	Liquid	Drizzle	Rain
P(Ze)	X1/dBZ	−5	−60	−40	−60	−25	−10
	X2/dBZ	0	−50	−17	−60	−17	5
	X3/dBZ	20	−10	5	−17	4	20
	X4/dBZ	25	0	10	−10	8	25
P(V)	X1/m·s−1	−0.5	−2	−2	−2	−2	−1
	X2/m·s−1	0	−1	−0.5	−2	0.5	2.5
	X3/m·s−1	2.5	1	2	0.5	2	10
	X4/m·s−1	8	3	4	1.5	4	10
P(W)	X1/m·s−1	0	0	0.1	0.1	0	0
	X2/m·s−1	0	0	0.4	0.4	0.5	1
	X3/m·s−1	4	0.4	4	2	3	3
	X4/m·s−1	4	0.6	4	3	4	4

); gave the coefficients of the affiliation function of Z, V, and, W for different hydrometeors; and selected the trapezoidal affiliation function to calculate the degree of affiliation for Z, V, and W as a way to determine the type of hydrometeor. The coefficients of the affiliation function of each radar parameter corresponding to each type of hydrometeor correspond to the range of thresholds corresponding to this radar parameter for that hydrometeor. Therefore, the membership function coefficients for each radar parameter corresponding to each hydrometeor are equivalent to the threshold range of this radar parameter for that hydrometeor type. In order to check the reasonableness of the classification results in Figure 5a, the various types of hydrometeors identified by the micro rain radar were matched with the corresponding millimeter-wave cloud radar observations at the same time and altitude layer, and the frequency distributions of the cloud radar albedo factor, Doppler velocity, and velocity spectral width corresponding to the various types of hydrometeors were statistically calculated (Figure 5b–d) and compared with the threshold ranges of Z, V, and W in Table 1. The results showed that the micro rain radar identified snow and mixed phase corresponding to the millimeter-wave cloud radar albedo factor, Doppler velocity, and velocity spectral width ranges are more consistent with the results in Table 1, but the ranges of Z and V thresholds corresponding to drizzle and rain are more different from those corresponding to drizzle and rain in Table 1.

The main reason for these discrepancies is that Garcia-Benadi et al. [25] differentiated between small-scale drizzle and large-scale raindrops using the skewness of the micro rain radar power spectrum, directly referencing the research results of Acquistapace et al. [17], who utilized the skewness of the millimeter-wave cloud radar power spectrum to distinguish between large-scale drizzle and small-scale cloud droplets. However, this approach did not account for the detection differences between micro rain radar and cloud radar equipment, necessitating adjustments to the Sk threshold values.

In response to the aforementioned issues, adjustments were made to the identification method for drizzle and raindrop thresholds. Sensitivity experiments were conducted by applying three sets of conditions:

• ① ${|\mathrm{S}}_{\mathrm{k}}|\ge 0.5$ and ${∆\mathrm{Z}}_{\mathrm{e}}\ge 1 \mathrm{dBZ}$;
• ② ${\mathrm{S}}_{\mathrm{k}}\ge -0.5$ and $1 \mathrm{dBZ}\le {∆\mathrm{Z}}_{\mathrm{e}}\le 2 \mathrm{dBZ}$;
• ③ ${\mathrm{S}}_{\mathrm{k}}\ge -0.5$ and $-{10 \mathrm{dBZ}\le ∆\mathrm{Z}}_{\mathrm{e}}\le 10 \mathrm{dBZ}$.

The sensitivity experiment results under the three aforementioned conditions were mapped to the cloud radar (CR), as illustrated in Figure 6.

The peak areas of cloud radar albedo factor, Doppler velocity, and velocity spectral width corresponding to the drizzle identified in condition ① are located on the right side of the raindrops (Figure 6a1,b1,c1), which is not in line with the actual situation and is also quite different from the ranges given by the Shupe et al. [10] study. The range of the distribution of the albedo factor, Doppler velocity, and velocity spectral width of the drizzle and raindrops under Condition ② still differed greatly from the results of Shupe (2007) et al. [10].

The drizzle and raindrops identified in condition ③ showed single-peak characteristics, with the drizzle albedo factor mainly concentrated in the range of −10–20 dBZ and the raindrop albedo factor mainly concentrated in the range of 5–30 dBZ. The Doppler velocities of the drizzle were mainly concentrated in the range of 3–6 m/s, and the widths of the velocity spectra were mainly concentrated in the range of 0.5–1.7 m/s. For the raindrops, the Doppler velocities were concentrated in the range of 2–9 m/s, and the widths of the velocity spectra were concentrated in the range of 0.3–1.7 m/s. The distribution of the drizzle and raindrop signals in this condition was more consistent with the results of Shupe et al.—0.3–1.7 m/s. The range of gross and raindrop signal distributions under this condition is more consistent with the findings of Shupe et al. [10].

Condition ③ was used to reclassify the LP and LD condensates, and the classification results were compared with the rain intensity observed by the surface precipitation phenomenometer at the corresponding times; the results are shown in Figure 7. Figure 7b2 shows that the rain intensity observed by the surface precipitation phenomenometer at the times of 14:30–14:50, 17:40–20:00, and 20:50–21:20 is larger, and the values are in the range of 0.3–4.7 mm/h. The corresponding hydrometeor classification results for these times are mainly raindrops, and the rain intensity corresponding to the times when drizzle is the main type of hydrometeor is less than 0.1 mm/h. The results of the LD and LP comparisons are similar (Figure 7a1,b1), which indicates that the modification of the hydrometeor classification method of Garcia-Benadi et al. [25] by utilizing Condition ③ is justified in this paper.

3.3. Differences in Hydrometeor Phase at Various Precipitation Stages

Based on the hydrometeor phase identification results in Figure 7 and the observations from the ground precipitation gauge, the precipitation process is divided into three stages: weak (stage 1), strong (stage 2), and weak (stage 3). A comparison is made of the percentages of hydrometeors at different heights during each stage. It is worth noting that during the precipitation process, cloud particle information not detected by the micro rain radar (MRR) can be obtained from the cloud radar (CR) [27].

During Stage 1, the onset of precipitation at the LD site preceded that at the LP site due to the northwest-to-southeast movement of the precipitation system, with LD, situated on the western side of the Liupan Mountains, being affected earlier than LP. A comparison between the two sites (Figure 8) reveals that at the mountaintop LP site, the proportion of raindrops is significantly lower compared to LD, with the former accounting for approximately 28% and the latter around 43%. This disparity may be attributed to the lower elevation of the LD site, located in a valley where lower-level moisture is more abundant compared to the mountaintop LP site. Consequently, in a more moisture-rich environment, drizzle at LD is more prone to coalescence and growth into larger raindrops. On the other hand, the difference in water vapor between the two sites can also be seen from the comparison of the percentage of cloud droplet-scale particles that can be detected by the millimeter-wave cloud radar but not by the micro rain radar; such small-scale particles accounted for about 5% of the particles in the LD, and only 3% of the particles appeared only in the range of 450 m above the ground in the LP, coupled with the LP being located on the top of the mountain, and the altitudinal layer of the raindrops below the zero layer grows thinner than that of the LD, so in the early stage of the precipitation, the rain droplet scale growth rate is slow, and the surface precipitation is also significantly less than LD.

In the Stage 2 period (Figure 9), as the precipitation system continued to advance from west to east, the water vapor within the altitude layer above the zero-degree layer was replenished, and the gradually decreasing percentage of cloud droplet-scale particles in the mixing layer of the two sites was reflected in the decrease in the altitude layer. The average percentage of the LP was as high as 8%, which further facilitated the growth of the ice and snow crystals in the clouds and led to more precipitation particles being produced at the altitude below the melting layer than in Stage 1. The ground-level precipitation reaches the maximum value of this precipitation.

Comparing the hydrometeor distributions of LP and LD during Stage 2, it is observed that the proportion of raindrops in LP remains lower than in LD, while the proportion of drizzle is higher than in LD. This similarity in distribution characteristics with Stage 1 is primarily attributed to the different elevations of the two sites. LP, being at a higher altitude with lower moisture content and a thinner layer below the melting level, prevents drizzle from growing sufficiently to raindrop size before reaching the ground, leading to lower ground precipitation amounts during this period compared to LD.

During Stage 3 (Figure 10), as this precipitation process enters the end, the water vapor content gradually decreases, and the small-scale drizzle lacks sufficient water vapor to further grow into larger-scale raindrops, so the two stations accounted for significantly fewer raindrops in this process than in Stage2. Due to the low elevation of the LD station and higher water vapor content than LP, about 6% of cloud droplet-scale particles are present in Stage 3, and the percentage of cloud droplet-scale particles tends to increase with decreasing altitude strata.

The LD raindrop percentage during Stage 3 is lower than that of the LP, which may be related to the influence of topography during precipitation. With the continuation of the precipitation system, the water vapor in the lower layers of the LD blocked by the mountain slopes is consumed during the strong precipitation period of Stage 2, while the mountain tops are constantly replenished by the warm and humid airflow in the lower layers lifted up by the topography, which provides a certain amount of water vapor conditions for the growth of the drizzle into raindrops [35].

4. Discussion

Micro rain radar and millimeter-wave cloud radar have different sensitivities for detecting hydrometeors at different scales, with the former mainly used for detecting millimeter-scale hydrometeors and the latter being more sensitive to sub-millimeter cloud droplet-scale particles. The RaProM algorithm used in this paper mainly uses the micro rain radar power spectrum information to identify the type of hydrometeor, but when setting the threshold for the micro rain radar power spectrum skewness, it applies the results of Acquistapace et al. [17], who used millimeter-wave cloud radar to study the furry rains, so the drizzle and raindrops identified directly by the RaProM algorithm have large deviations from the ground-based raindrop spectrum observations. The sensitivity experiments with the RaProM algorithm provided a way to determine the thresholds for parameters such as skewness. After adjusting the parameter thresholds in the modified algorithm, the identification results for drizzle and raindrops showed good consistency with actual observations.

On the basis of the above work, by comparing the micro rain radar observation data and hydrometeor identification results of the valleys and mountain tops on the west side of Liupan Mountain, it is found that the topography of Liupan Mountain has a greater influence on the distribution of hydrometeors during precipitation, mainly reflected in the following: at the beginning of precipitation, because the valleys are more abundant in water vapor than the mountain tops, they can produce more large-scale raindrops than the mountain tops. When precipitation develops to a stronger period, because of the high altitude and low water vapor content of the mountain tops, the altitude layer below the melting layer is thin, and the drizzle is not able to grow sufficiently to the raindrop scale before it settles to the ground, thus making the surface precipitation in this period less than that in the valley. The percentage of raindrops in the valleys is lower than that on the mountain tops during the transition to weak precipitation until the end of the precipitation period, which is related to the fact that the valleys are depleted of water vapor after the period of strong precipitation, so that the mountain tops have a water vapor environment that produces more raindrops than the valleys during the period of the end of the precipitation.

5. Conclusions

This study utilizes observations from a micro rain radar to investigate a stratiform cloud precipitation event during the rainy season in the Liupan Mountains. Simultaneously, it combines co-observations from a millimeter-wave cloud radar and a ground-based precipitation gauge to study the evolution characteristics of different types of hydrometeors during the precipitation process. The primary research findings are as follows:

• The distribution periods of drizzle and raindrops classified using the RaProM method reveal discrepancies between the observed raindrop spectra and rainfall intensity from the DSG5 ground-based precipitation gauge. Assimilating the classified hydrometeors into millimeter-wave cloud radar data for the same height and time intervals, significant differences are found in the radar reflectivity, velocity, and velocity spectrum width thresholds corresponding to drizzle and raindrops compared to the existing observations.
• After adjusting the skewness parameter and differential albedo factor parameter in the RaProM method, sensitivity experiments were conducted to select the parameter thresholds that are more consistent with the ground-based precipitation phenomenometer observations and reclassify the hydrometeors of this precipitation process; the corresponding threshold ranges of cloud radar albedo factors, velocities, and velocity spectral widths of the classified hydrometeor are in good agreement with the existing observations.
• During the three phases of this precipitation process—weak, strong, and weak—there were obvious evolutionary characteristics of various hydrometeors. The hydrometeor phase above the zero-degree layer mainly existed in the form of snow; in the weak precipitation stage, the hydrometeor phase below the zero-degree layer mainly existed in the form of furry rain; and in the strong precipitation stage, it existed in the form of raindrops. During the weak precipitation phase, the percentage of cloud droplet-scale particles at the bottom site was higher than that at the top site. During the strong precipitation stage, the percentage of raindrops at the bottom site was higher than that at the top site. The percentage of raindrops after the strong precipitation period was higher at the summit site than at the bottom site. Topography plays an important role in the evolution of raindrops.